OpenAI says 75% of customers use AI for new tasks

OpenAI said that 75% of its customers reported using AI to complete tasks they had not previously been able to do.

The company noted that organizations have deployed ChatGPT for writing, coding, research, data analysis, and design.

OpenAI also said developers are using Codex to speed up their workflow and tackle more complex challenges.

These developments reflect a shift from experimentation to broader integration of AI in workplace processes, OpenAI said.

🔗 Source: OpenAI

🧠 Food for thought

Implications, context, and why it matters.

OpenAI says 75% of customers use AI for previously impossible tasks, yet it withholds counts of enterprise customers and the split of 3 million paying business users across ChatGPT Enterprise (OpenAI’s enterprise-grade tier) versus lower tiers 1. Without total customer numbers or an Annual Recurring Revenue (ARR) mix by segment, it’s unclear whether the 75% covers 100 firms or 100,000.
Enterprise solutions contribute 25-30% of OpenAI’s $12 billion ARR while consumer subscriptions deliver 55-60% 1.

General-purpose AI model providers (companies that build models for many use cases) face obligations starting August 2, 2025 2 3. Requirements span technical documentation; transparency reports; training data summaries; plus copyright compliance policies. Companies that substantially modify models through fine-tuning (retraining a model on domain- or company-specific data) become providers with full compliance duties 2. That spurs demand for governance solutions (tools for AI risk, compliance, audit) that track model lineage (the history of model versions and modifications) and automate reports.
Enterprise software vendors and AI infrastructure providers hit a milestone on August 2, 2026 when high-risk AI system rules start, so deployers (organizations that implement or use AI systems) must keep AI inventories (catalogs of AI systems in use) with monitoring 2 4. Vendors can time a 12-month sales cycle to match it, featuring synthetic content labeling (clearly marking AI-generated outputs) 4 and accessibility requirements (ensuring outputs are usable by people with disabilities) 4.